I only have detections closer than 1.5 m and beyond 4 meteres. And no detections between these two distances.

[filipetrocadoferreira]

I'm using Kinect 2:

and the config:

=====================================

Distance Range Accepted Detections

=====================================

distance_range_accepted_detections = 7 # up to what distance detections are kept. All detections after that distance are rejected

=====================================

ROI

=====================================

inc_width_ratio = -0.25 inc_height_ratio = -0.25 region_size_threshold = 20

======================

Freespace

======================

parameteters for occupancy Map computation

these parameter define the number of bins for the occupancy histogram

freespace_scaleZ = 20; freespace_scaleX = 20;

freespace_minX = -6;

freespace_minZ = 0;

freespace_maxX = 6;

freespace_maxZ = 7;

freespace_threshold = 500

freespace_max_depth_to_cons = 7

freespace_minX = -5; freespace_minZ = 0; freespace_maxX = 5; freespace_maxZ = 7; freespace_threshold = 170 freespace_max_depth_to_cons = 4.5

====================================

Detector

====================================

parameters upper body detector

evaluation_NMS_threshold_LM = 0.8 # lowering this parameter will reduce the number of false positives! Keep in mind this is distance! evaluation_inc_height_ratio = 0.3 # parameter for increasing the size of ROI which is then scanned by the depth template

<<< these parameters are only relevant if you want to perform multi-scaling, evaluation_nr_scales > 1

evaluation_stride = 3 evaluation_scale_stride = 1.03 evaluation_nr_scales = 1 evaluation_inc_cropped_height = 20 evaluation_greedy_NMS_overlap_threshold = 0.3 evaluation_greedy_NMS_threshold = 0.25

>>>

scanning corridor, which means for each ROI we scan it with the depth template starting from min_height (in meters) for up to max_height

max_height = 2.2 min_height = 1.4

======================

World scale

======================

WORLD_SCALE = 1.0

===========================

height and width of images

===========================

dImWidth = 320 dImHeight = 270

====================================

Number of Frames / offset

====================================

numberFrames = 1000000 nOffset = 0

====================================

Size of Template

====================================

template_size = 30

[filipetrocadoferreira]

In my opinion one of the causes can be:

In Detector::ComputeFreespace:

occ_map(pos_x, pos_z) += z;

Closer objects will have of course an heavy weight on the occupancy map and with the distance, even with small number of points, further objects will have a big weight, but maybe this will cause a gap between some distances as I see in my case.